Interconnects: open-model inference volume shows near-exponential growth, led by Fireworks
2026-11-04
Nathan Lambert's Interconnects reports that open-model inference startups (Fireworks, Baseten, Together AI) and OpenRouter's open-model token share show near-exponential token-volume growth, with an uptick following GLM 5.2's agentic release. Fireworks disclosed at its first company conference that it now serves more than 40 trillion tokens per day to customers' customized models in one vertical that grew over 2x since summer, implying a total figure likely well above 100T tokens/day by this fall.
Significance 3: A credible researcher's analysis citing public company disclosures quantifies real growth in open-model commercial adoption, a meaningful strategy signal though not an independent benchmark or verdict-moving result.
Implications · machine-drafted, not owner judgment
If open-model inference volume is really compounding this fast, independent inference providers are becoming a meaningful commercial layer in their own right, not just a cost-sensitive stopgap — strengthening the practical case for open-weights adoption in enterprise and local-build contexts, separate from whether open models remain 'one generation behind' on raw capability.
- Whether Fireworks or a peer publishes a comparable token-volume figure next quarter confirming the growth rate.
- Whether OpenRouter's open-model token share rises or plateaus in its next public report.
Sources
- curator How big is the open model economy today? retrieved 2026-11-04